Mistral OCR 3 has achieved a 74% overall win rate compared to its predecessor, Mistral OCR 2. This improvement is particularly notable in its handling of forms, scanned documents, complex tables, and handwriting. The new model outperforms both traditional enterprise document processing solutions and AI-based OCR systems, making it a strong contender in the market. The model extracts text and images from various document types with high accuracy and supports markdown output that incorporates HTML-based table reconstruction. This feature allows downstream systems to interpret both content and structure effectively. Priced at $2 per 1,000 pages, with a 50% discount for batch processing, it offers a competitive cost advantage. Developers can access Mistral OCR 3 via API, while the Document AI Playground provides a user-friendly interface for parsing documents into clean text or structured JSON. Mistral OCR 3 is designed for versatility, excelling in areas where many OCR solutions struggle. It accurately processes handwritten text, detecting cursive and layered annotations, as well as managing complex forms with intricate layouts like invoices and compliance documents. Its robustness against low-quality scans and compression artifacts further enhances its reliability for real-world applications.